Rework ecp_mod_p192()
On x86_64, this makes it 5x faster, and ecp_mul() 17% faster for this curve.
The code is shorter too.
diff --git a/library/ecp.c b/library/ecp.c
index d3880be..d53d306 100644
--- a/library/ecp.c
+++ b/library/ecp.c
@@ -475,25 +475,36 @@
}
#if defined(POLARSSL_ECP_DP_SECP192R1_ENABLED)
-/*
- * 192 bits in terms of t_uint
- */
-#define P192_SIZE_INT ( 192 / CHAR_BIT / sizeof( t_uint ) )
-/*
- * Table to get S1, S2, S3 of FIPS 186-3 D.2.1:
- * -1 means let this chunk be 0
- * a positive value i means A_i.
- */
-#define P192_CHUNKS 3
-#define P192_CHUNK_CHAR ( 64 / CHAR_BIT )
-#define P192_CHUNK_INT ( P192_CHUNK_CHAR / sizeof( t_uint ) )
+/* Add 64-bit chunks (dst += src) and update carry */
+static inline void add_64( t_uint *dst, t_uint *src, t_uint *carry )
+{
+ unsigned char i;
+ t_uint c = 0;
+ for( i = 0; i < 8 / sizeof( t_uint ); i++, dst++, src++ )
+ {
+ *dst += c; c = ( *dst < c );
+ *dst += *src; c += ( *dst < *src );
+ }
+ *carry += c;
+}
-const signed char p192_tbl[][P192_CHUNKS] = {
- { -1, 3, 3 }, /* S1 */
- { 4, 4, -1 }, /* S2 */
- { 5, 5, 5 }, /* S3 */
-};
+/* Add carry to a 64-bit chunk and update carry */
+static inline void carry64( t_uint *dst, t_uint *carry )
+{
+ unsigned char i;
+ for( i = 0; i < 8 / sizeof( t_uint ); i++, dst++ )
+ {
+ *dst += *carry;
+ *carry = ( *dst < *carry );
+ }
+}
+
+#define OFFSET ( 8 / sizeof( t_uint ) )
+#define A( i ) ( N->p + ( i ) * OFFSET )
+#define ADD( i ) add_64( p, A( i ), &c )
+#define NEXT p += OFFSET; carry64( p, &c )
+#define LAST p += OFFSET; *p = c; while( ++p < end ) *p = 0
/*
* Fast quasi-reduction modulo p192 (FIPS 186-3 D.2.1)
@@ -501,53 +512,27 @@
static int ecp_mod_p192( mpi *N )
{
int ret;
- unsigned char i, j, offset;
- signed char chunk;
- mpi tmp, acc;
- t_uint tmp_p[P192_SIZE_INT], acc_p[P192_SIZE_INT + 1];
+ t_uint c = 0;
+ t_uint *p, *end;
- tmp.s = 1;
- tmp.n = sizeof( tmp_p ) / sizeof( tmp_p[0] );
- tmp.p = tmp_p;
+ /* Make sure we have the correct number of blocks */
+ MPI_CHK( mpi_grow( N, 6 * OFFSET ) );
+ p = N->p;
+ end = p + N->n;
- acc.s = 1;
- acc.n = sizeof( acc_p ) / sizeof( acc_p[0] );
- acc.p = acc_p;
-
- MPI_CHK( mpi_grow( N, P192_SIZE_INT * 2 ) );
-
- /*
- * acc = T
- */
- memset( acc_p, 0, sizeof( acc_p ) );
- memcpy( acc_p, N->p, P192_CHUNK_CHAR * P192_CHUNKS );
-
- for( i = 0; i < sizeof( p192_tbl ) / sizeof( p192_tbl[0] ); i++)
- {
- /*
- * tmp = S_i
- */
- memset( tmp_p, 0, sizeof( tmp_p ) );
- for( j = 0, offset = P192_CHUNKS - 1; j < P192_CHUNKS; j++, offset-- )
- {
- chunk = p192_tbl[i][j];
- if( chunk >= 0 )
- memcpy( tmp_p + offset * P192_CHUNK_INT,
- N->p + chunk * P192_CHUNK_INT,
- P192_CHUNK_CHAR );
- }
-
- /*
- * acc += tmp
- */
- MPI_CHK( mpi_add_abs( &acc, &acc, &tmp ) );
- }
-
- MPI_CHK( mpi_copy( N, &acc ) );
+ ADD( 3 ); ADD( 5 ); NEXT; // A0 += A3 + A5
+ ADD( 3 ); ADD( 4 ); ADD( 5 ); NEXT; // A1 += A3 + A4 + A5
+ ADD( 4 ); ADD( 5 ); LAST; // A2 += A4 + A5
cleanup:
return( ret );
}
+
+#undef OFFSET
+#undef A
+#undef ADD
+#undef NEXT
+#undef LAST
#endif /* POLARSSL_ECP_DP_SECP192R1_ENABLED */
#if defined(POLARSSL_ECP_DP_SECP521R1_ENABLED)
diff --git a/tests/suites/test_suite_ecp.data b/tests/suites/test_suite_ecp.data
index 9eb302b..4748ff9 100644
--- a/tests/suites/test_suite_ecp.data
+++ b/tests/suites/test_suite_ecp.data
@@ -253,14 +253,26 @@
depends_on:POLARSSL_ECP_DP_SECP192R1_ENABLED
ecp_gen_keypair:POLARSSL_ECP_DP_SECP192R1
+ECP mod p192 small (more than 192 bits, less limbs than 2 * 192 bits)
+depends_on:POLARSSL_ECP_DP_SECP192R1_ENABLED
+ecp_fast_mod:POLARSSL_ECP_DP_SECP192R1:"0100000000000103010000000000010201000000000001010100000000000100"
+
ECP mod p192 readable
depends_on:POLARSSL_ECP_DP_SECP192R1_ENABLED
-ecp_fast_mod:POLARSSL_ECP_DP_SECP192R1:"000000000000010500000000000001040000000000000103000000000000010200000000000001010000000000000100"
+ecp_fast_mod:POLARSSL_ECP_DP_SECP192R1:"010000000000010501000000000001040100000000000103010000000000010201000000000001010100000000000100"
+
+ECP mod p192 readable with carry
+depends_on:POLARSSL_ECP_DP_SECP192R1_ENABLED
+ecp_fast_mod:POLARSSL_ECP_DP_SECP192R1:"FF00000000010500FF00000000010400FF00000000010300FF00000000010200FF00000000010100FF00000000010000"
ECP mod p192 random
depends_on:POLARSSL_ECP_DP_SECP192R1_ENABLED
ecp_fast_mod:POLARSSL_ECP_DP_SECP192R1:"36CF96B45D706A0954D89E52CE5F38517A2270E0175849B6F3740151D238CCABEF921437E475881D83BB69E4AA258EBD"
+ECP mod p192 (from a past failure case)
+depends_on:POLARSSL_ECP_DP_SECP192R1_ENABLED
+ecp_fast_mod:POLARSSL_ECP_DP_SECP192R1:"1AC2D6F96A2A425E9DD1776DD8368D4BBC86BF4964E79FEA713583BF948BBEFF0939F96FB19EC48C585BDA6A2D35C750"
+
ECP test vectors secp192r1 rfc 5114
depends_on:POLARSSL_ECP_DP_SECP192R1_ENABLED
ecp_test_vect:POLARSSL_ECP_DP_SECP192R1:"323FA3169D8E9C6593F59476BC142000AB5BE0E249C43426":"CD46489ECFD6C105E7B3D32566E2B122E249ABAADD870612":"68887B4877DF51DD4DC3D6FD11F0A26F8FD3844317916E9A":"631F95BB4A67632C9C476EEE9AB695AB240A0499307FCF62":"519A121680E0045466BA21DF2EEE47F5973B500577EF13D5":"FF613AB4D64CEE3A20875BDB10F953F6B30CA072C60AA57F":"AD420182633F8526BFE954ACDA376F05E5FF4F837F54FEBE":"4371545ED772A59741D0EDA32C671112B7FDDD51461FCF32"